Urban environments often bear the brunt of rapid growth and climate change effects, leading cities worldwide to grapple with how effectively they can manage and expand their green spaces. A new development from researchers at New York University (NYU) aims to change how urban planners approach this challenge by employing advanced artificial intelligence (AI) and satellite technology.
The core of this innovation is led by Rumi Chunara, an associate professor at NYU with dual appointments at the Tandon School of Engineering and the School of Global Public Health. Chunara's team unveiled their cutting-edge AI system, which utilizes high-resolution satellite imagery to track urban green spaces with unprecedented accuracy, particularly within densely populated areas like Karachi, Pakistan.
Karachi served as the ideal testing ground for this technology, as it embodies the need for precise measurements related to urban vegetation against the backdrop of its diverse urban conditions. The findings are significant: the team’s research, now accepted for publication by the ACM Journal on Computing and Sustainable Societies, exposed stark environmental divides across the city. While some neighborhoods benefit from tree-lined streets, others exist nearly devoid of vegetation.
Cities have traditionally faced challenges tracking their green spaces, from parks to individual street trees. Existing methods often miss up to 37% of urban vegetation, which is concerning, especially as cities contend with rapid urbanization and climate change. Accurate measurements of green spaces have become increasingly important not just for aesthetic reasons but also for public health and environmental management, as these areas help lower urban temperatures, filter air pollutants, and serve as venues for physical and mental wellness.
Chunara's innovative approach involved the enhancement of AI segmentation architectures, particularly using the DeepLabV3+ model, to significantly boost the accuracy of urban green space mapping. The researchers employed high-resolution satellite imagery from Google Earth to train their system, taking the novel step of what they termed 'green augmentation'—training data was enhanced to include different versions of green vegetation captured under varying lighting and seasonal conditions.
This methodological shift yielded impressive results: the AI model achieved 89.4% accuracy and 90.6% reliability compared to traditional methods, which reported only 63.3% accuracy. Chunara emphasizes the necessity of this new data, stating, "Our system learns to recognize more subtle patterns... this type of data is necessary for urban planners to identify neighborhoods..."
The analysis of Karachi's urban green spaces revealed not only the need for improved city planning but also stark disparities across neighborhoods. On average, residents of Karachi have access to just 4.17 square meters of green space per person, which is less than half of the World Health Organization's recommended minimum of 9 square meters. The research revealed protestingly uneven access to green amenities, with some outlying union councils reporting over 80 square meters per person, whereas other areas barely register at 0.1 square meters.
One notable discovery from the research was the correlation between vegetation and urban temperatures. The study showed clear evidence: areas with more vegetation were cooler compared to those with higher amounts of paved surface. Chunara pointed out the importance of accurately mapping these green resources, noting, "Without accurate mapping, cities cannot address disparities effectively." This finding aligns well with successful urban practices seen elsewhere; for example, Singapore, with similar population density to Karachi, provides nearly 9.9 square meters of green space per person, effectively managing urban greenery through planning and policy.
The researchers have made their methodology publicly available, though they caution potential users: applying the AI system to other cities would necessitate retraining with local satellite imagery. This study falls within the broader framework of Chunara's research interests, which include exploring methods to understand variances in public health based on infrastructural and environmental discrepancies. Previous investigations have involved data mining to assess issues like systemic racism and mental health's interconnected nature.
Funding for this significant research came from the National Science Foundation and the National Institutes of Health, highlighting the importance seen at the institutional level for addressing environmental health gaps. Chunara's work demonstrates how data science can shine light on underrepresented areas within urban planning and public health discussions.
Overall, as cities continue to grow and struggle with environmental impacts, methodologies like the one developed by Chunara and his team may be pivotal for ensuring equitable access to green spaces, promoting healthier urban livelihoods, and enhancing the quality of life for residents across various socioeconomic backgrounds.